glass

glass

Give your coding agent hands for the app it's building — launch it, drive it, and verify the result, without burning a screenshot on every step.

A Rust MCP server that gives an AI coding agent a closed build → see → interact → debug loop over external native GUI applications.

glass lets an agent launch a GUI app, capture what is on screen, inject mouse and keyboard input, read the app's logs, and detect visual changes — so a coding agent can build and debug UI applications independently instead of asking the user "does this look right?".

glass drives apps as an external black box, so it works with any native GUI app regardless of toolkit or language. It has two Linux backends (X11 and Wayland), a Windows backend, an Android backend (an AVD emulator, driven over adb from any host), an iOS backend (native apps in the Simulator over xcrun simctl, with input and the accessibility tree via idb_companion, including a two-finger pinch), and a macOS backend, behind a platform-agnostic core.

The same input tools work across backends. Default Linux displays and mobile emulators keep input separate from your desktop. Windows shares it; macOS adds a preview of separate-account setup for apps off your desktop, alongside its existing shared-desktop setup. glass_capabilities with scope: "session" reports click, scroll and text support and whether each can affect your desktop.

The macOS account setup uses the normal MCP tools and one Glass installation. Application setup… guides enrollment, permission handoff and live readiness checks, including creating a standard development account with a locally chosen password. It needs a real login and Glass privacy grants in that account, retained between runs. Enrollment applies to your macOS user; glass-mcp unenroll returns to shared-desktop use while retaining the app, development account and MCP startup integration. Development loops have been tested on an Apple Silicon Mac mini without a physical monitor. These runtime checks used an initialized development account and include physical keyboard/mouse coexistence, recovery after an interrupted client, and starting its development desktop after a real reboot with existing permissions. New-account onboarding has also reached Ready and passed ordinary app interaction on that Mac. Initial macOS login and any missing privacy grants require user action; setup on a clean macOS installation without prior Glass grants has not been tested. Application setup includes Start environment… for a missing development desktop, using local account credentials and an already enabled macOS Screen Sharing service. This guided post-reboot flow and ordinary build/interact/rebuild/stop loops in an indexed workspace have been tested without repeating enrollment or permission setup. Unattended development-desktop startup is not implemented.

See it

An agent building a GUI app runs it under glass, reproduces a bug from the accessibility tree (no screenshots), fixes the code, and re-verifies — the loop it otherwise can't close on its own. Try it yourself.

Try it in 60 seconds

1. Download glass for your platform from the Releases page and connect it to your agent — glass works with any MCP host (see which are verified).
2. Get the example app — clone this repo, or download examples/tasks-demo/tasks_demo.py (on Linux it needs sudo apt install python3-gi gir1.2-gtk-4.0).
3. Paste this to your agent:

Use glass to run examples/tasks-demo/tasks_demo.py with accessibility on. There's a bug: clicking

Add doesn't add the typed task. Reproduce it by driving the UI and checking the

accessibility tree (don't just screenshot), then find and fix the bug in the code and verify a

task actually appears.

Your agent launches the app, reproduces the bug from the accessibility tree, fixes the one-line wiring bug, and confirms the task appears — the whole build → see → interact → debug loop, start to finish. (glass-mcp doctor checks your environment if anything's off.)

The loop in practice

An agent can fill in a form, click Save, and check that the app reports success. When the app exposes an accessibility tree, the agent works with named controls and verifies changes from text:

glass_start { "run": ["python3", "app.py"] }

glass_do { "actions": [ { "action": "set_value", "target": { "query": "account", "role": "TextField", "states": ["enabled"] }, "text": "hello" }, { "action": "click_element", "target": { "query": "Save", "role": "Button", "states": ["enabled"] }, "mode": "auto" }, { "action": "wait_for_element", "name": "Saved" } ] } glass_logs

Use glass_find_elements to inspect candidates when the intended target is not unique or known well enough to act on directly. Use glass_a11y_snapshot to explore the app's controls. See the tool reference for click modes and platform behavior.

For a canvas or custom-rendered app with no accessibility tree, drive it by pixels instead — glass_screenshot, glass_click {x,y}, and glass_diff, which returns changed_pct + a bbox as text, so routine checks between screenshots cost no vision tokens. Why the loop is shaped this way: the build → see → interact → debug loop.

To review a run after the app stops, opt into session evidence recording with --trace-dir. Glass retains tool inputs and requested results in bounded storage, then glass-mcp trace inspect and trace export validate the trace and package its evidence in a ZIP. Recording is off by default; retained app content can contain sensitive data.

Install at a glance

Download the latest build for your platform from the Releases page, then set up your host:

• Linux — docs/how-to/setup-linux.md (X11 or Wayland; Xvfb / sway + bubblewrap)
• Windows — docs/how-to/setup-windows.md (a prebuilt .exe + Sandboxie)
• macOS — docs/how-to/setup-macos.md (install the notarized .dmg; no build needed)
• Android — docs/how-to/setup-android.md (an AVD emulator, from any host)
• iOS — docs/how-to/setup-ios.md (the Simulator, macOS host only)

Every asset is listed in docs/reference/platforms.md. Prefer to compile, or on an architecture with no published asset? See docs/how-to/build-from-source.md — it is a single cargo build.

Then connect glass to your agent and run glass-mcp doctor to check the environment. New here? Follow the tutorial for a guaranteed first success.

The full toolbox remains the default. For fewer agent tool definitions, start with glass-mcp --tool-profile lean and use glass_do for actions. Inspect either inventory without starting a session using glass-mcp tools --json; see tool profiles.

Drive it well — the glass-drive skill

glass needs no app integration and no skill to run, but an agent drives it far more reliably with the open glass-drive Agent Skill — it stops the agent spending its first turns rediscovering the verify-cheaply-then-look loop. Installing it is the single highest-leverage thing you can add when pointing an agent at glass.

Platform support

✓ supported · ◑ partial · – not supported · 🚧 planned.

Full matrix, per-capability detail, and system requirements: docs/reference/platforms.md. Transport is MCP over stdio (default) or network HTTP.

Documentation

The full docs — tutorial, how-to guides, reference, and explanations — are under [`docs/`](docs/README.md). See CHANGELOG.md for release notes, and Stability and versioning for what a 1.0 release guarantees.

Contributing? CONTRIBUTING.md has the gates a PR has to pass, and Verify a change covers platform code your own host cannot run.

License

glass is open core, licensed Apache-2.0 — see LICENSE-APACHE.